Related Experiment Videos
Weight changes and all-cause mortality in critically ill patients: a multi-center retrospective cohort study
Cheng Luo1, Tinglong Zhang2, Xiaoyong Xiao1
1Department of Emergency Medicine, Shenzhen Second People's Hospital, The First Affiliated Hospital of Shenzhen University, Shenzhen, China.
Objective:
Body weight changes is a readily available composite metric in critically ill patients that may reflect both fluid balance and nutritional-metabolic status. However, large-scale evidence on its independent association with-and predictive value for-patient-centered outcomes such as mortality remains limited.
Methods:
This large, multicenter retrospective study used the U.S. eICU Collaborative Research Database. A total of 30,537 adult critically ill patients were included. The primary exposure was weight change rate. Feature selection was performed with the Boruta machine-learning algorithm. Multivariable logistic regression was used to estimate the association between weight change rate and mortality. Non-linear relationships and threshold effects were explored with generalized additive models (GAM) and two-piece linear regression. Discriminative performance was assessed with the area under the receiver operating characteristic curve (AUC).
Results:
Boruta identified body weight change rate as an important predictor of both ICU and hospital mortality. After multivariable adjustment, each additional percentage of weight gain was associated with a significant increase in ICU mortality (OR 1.04, 95% CI 1.03-1.05) and hospital mortality (OR 1.03, 95% CI 1.03-1.04) (both p < 0.0001). The association between weight change rate and mortality risk was non-linear, with a slope change (threshold effect) occurring at approximately 5% (ICU mortality: 5.14%; hospital mortality: 4.73%). Weight change rate provided superior discrimination compared with static indices such as admission weight, discharge weight, or admission BMI.
Conclusion:
In this large cohort, the ICU-derived weight change rate is an independent predictor of mortality in critically ill patients and exhibits a non-linear relationship with a threshold effect at approximately 5%. Dynamic monitoring of weight change rate outperforms static anthropometric measures and may serve as a simple yet powerful bedside tool for prognostic assessment.
